On the representational nature of grammatical gender in early and late bilinguals: A test for the Gender Acquisition and Processing (GAP) hypothesis